Pith. sign in

REVIEW 4 major objections 6 minor 45 references

Trace-of-Thought Prompting: Investigating Prompt-Based Knowledge Distillation Through Question Decomposition

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Trace-of-Thought prompting distills arithmetic reasoning by having a teacher decompose each problem into steps that a smaller student follows, lifting small-model GSM8K accuracy by up to 113%.

desk verdict A training-free two-stage decomposition prompt that shows real GSM8K gains for small models, but the paper's own examples and selective statistics leave the core distillation claim unverified; deserves peer review with a mandatory leakage audit and full reporting. read the letter →

arxiv 2504.20946 v2 pith:MZGYOO24 submitted 2025-04-29 cs.CL cs.AI

Tyler McDonald , Ali Emami This is my paper · ORCID
classification cs.CLcs.AI
keywords Trace-of-ThoughtPromptingprompt-basedknowledgedistillationquestiondecompositionarithmeticreasoningin-contextlearningGSM8KMATHlow-resourcelanguagemodels
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper proposes Trace-of-Thought Prompting, a prompt-only form of knowledge distillation in which a large teacher model first breaks an arithmetic problem into short step-by-step prompts and a smaller student model then solves the original problem using those steps. The authors argue that this transfers reasoning capability without fine-tuning, and they report accuracy gains over the best alternative prompt of up to 113% on GSM8K and roughly 21% on MATH, concentrated in low-resource models such as Llama 2 and Zephyr. The flagship result is Llama 2 7B rising from 23.5% to 50% on GSM8K when GPT-4 supplies the steps. The paper also finds that a modest 8B model, Llama 3, can serve as teacher and still improve weaker students, pointing toward open-source models as both students and teachers.

What carries the argument

The carrying mechanism is the two-stage prompt pair: a delegation template ('Create very short step-by-step prompts for the following problem... Format as a list. Do not solve the problem.') and a solution template ('We are given the following problem... Use the following steps to solve the problem: <steps>.'). The delegation stage turns the teacher into a source of decomposed reasoning traces, and the solution stage makes the student condition on those traces through in-context learning rather than weight updates. The paper formalizes this as $L_T(q) \to \{s_1,\ldots,s_n\}$ followed by $L_S(q \mid \{s\}) \to a$, with the claim that the intermediate steps carry the reasoning to be transferred.

What would settle it

Inspect the teacher's step lists across the test set and check whether successful student answers can be obtained by reading the final numeric value or a near-complete arithmetic expression out of the steps; a concrete version is to rerun the delegation prompt, mask or paraphrase every step that contains a number within one arithmetic operation of the final answer, and see whether Llama 2's GSM8K score stays near 50%.

Watch

Extended reading notes

Core claim

Trace-of-Thought Prompting is a two-phase protocol: in the delegation phase, the teacher receives 'Create very short step-by-step prompts for the following problem... Format as a list. Do not solve the problem'; in the solution phase, the student receives the original question plus those steps and is asked to use them to solve. On 200-question samples from GSM8K and MATH, the protocol outperforms standard, Chain-of-Thought, and Plan & Solve prompts for most low-resource students when GPT-4 is the teacher: Llama 2 7B improves from 23.5% to 50% on GSM8K and Zephyr from 30% to 55%, with both differences significant at $\alpha = 0.05$. With Llama 3 8B as teacher, Llama 2 and Zephyr still improve on GSM8K, reaching 37.5% and 43%. The authors frame this as prompt-based knowledge distillation: the teacher supplies key insights rather than explicit answers, and the student performs the calculation.

Load-bearing premise

The reported gains assume the teacher's generated steps do not contain the answer or near-answer arithmetic, so that the student's improvement reflects distilled reasoning rather than copying numbers out of the prompt; the paper itself flags that a too-strong delegation prompt can encourage the teacher to solve the problem and thereby contaminate the data.

Editorial extensions

If this is right

  • Student models up to 8 billion parameters can improve arithmetic accuracy by more than 100% relative to the best baseline prompts when a large teacher supplies steps, with no fine-tuning or gradient updates.
  • Low-resource models can act as teachers: Llama 3 8B steps produce significant GSM8K gains for Llama 2 and Zephyr, widening the pool of usable teacher models.
  • Because the entire pipeline is prompt-based, a user can apply it to a new task by swapping the teacher or changing the decomposition request, without model retraining.
  • The explicit step list makes the reasoning inspectable before the student executes it, so erroneous or suboptimal teacher steps can be caught and corrected in a human-in-the-loop process.
  • The gains concentrate in models with lower absolute performance, suggesting the method helps close the reasoning gap for weak reasoners rather than uniformly scaling all models.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If solution leakage is controlled, the same two-stage protocol could be used to generate fine-tuning corpora, converting prompt-based distillation into weight-based distillation by training students on teacher step lists paired with correct answers.
  • The framework should be testable on non-arithmetic decomposition tasks; the paper itself leaves abstract reasoning datasets open, so a natural extension is whether structurally similar step prompts transfer to pattern recognition or commonsense reasoning without leaking the answer.
  • Human-in-the-loop correction of erroneous teacher steps, such as the pension example in the qualitative analysis, could be turned into an interactive tutoring procedure where a user edits steps before the student solves, which the paper gestures at but does not evaluate systematically.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper introduces Trace-of-Thought (ToT) Prompting, a two-stage prompting method in which a teacher LLM decomposes a math word problem into short sub-prompts and a student LLM solves the original question using those steps. The authors evaluate zero-shot ToT against standard, chain-of-thought, and plan-and-solve prompting on 200 GSM8K and 200 MATH problems across six models, with GPT-4 and Llama-3-8B as teachers. They report large relative accuracy gains for small models (e.g., Llama-2-7B on GSM8K from 23.5% to 50% with GPT-4 teacher) and claim up to 113% gain on GSM8K and 21% on MATH, alongside qualitative examples of transparency and error analysis. The paper positions the method as a fine-tuning-free, prompt-based knowledge-distillation framework.

Significance. If the empirical results hold, the method is a useful, low-cost alternative to fine-tuning-based distillation for arithmetic reasoning, and the transparency of the delegation steps is a genuine practical asset. The paper introduces no fitted parameters or formal derivations, so the distillation claim is entirely empirical; its central evidentiary weight rests on the GSM8K z-tests with the GPT-4 teacher, four of which are significant at conventional levels. I credit the authors for releasing evaluation data, performing human annotation, and running significance tests; these are appropriate steps for a prompting-methods paper. However, the current reporting overstates MATH results and leaves the reasoning-versus-copy-out mechanism unverified, as detailed below.

major comments (4)
  1. [Abstract and Section 6.3 (Tables 3 and 4)] The claim of “21% on MATH” is not supported by any statistically significant result. The only MATH row in Tables 8 and 10 with a positive z near that magnitude is Zephyr with the low-resource teacher (14.5 vs 12, p = 0.60), and all MATH comparisons have p > 0.42. The abstract should either remove the MATH figure or state it as a descriptive relative gain with explicit notice of its non-significance.
  2. [Section 8.2 and Limitations (“Solution diffusion from delegation model”)] The central claim that ToT transfers reasoning rather than answers is undermined by answer leakage in the delegation output. In the Section 8.2 example, the teacher’s step “Acknowledge the function’s constant value: ? = 3” is the answer itself, and the Limitations section concedes that “too strong of a prompt may encourage the teacher model to solve the problem … leading to inadvertent data contamination.” The paper reports no audit of how often the 200 GSM8K (or MATH) delegation outputs contain the final answer or answer-determining arithmetic. Without such an audit, the flagship Llama-2 gain from 23.5% to 50% cannot be attributed to distilled reasoning rather than to copying leaked numbers. I therefore request a systematic leakage analysis, including examples and ideally accuracy on the subset of items with no leaked answer.
  3. [Appendix Tables 7–10] The significance testing is selective and uncorrected. The tables report p-values only for positive z, yet the captions say “Only scenarios with positive Z (gains) are reported” while also showing negative z as em dashes; more substantively, no adjustment is made for the multiple comparisons across six models and two teacher conditions. The marginal GSM8K gain for GPT-3.5-Turbo (p = 0.0477) would not survive a Bonferroni correction for six comparisons. Please report all comparisons, two-sided tests, and adjusted p-values or confidence intervals for the accuracy differences.
  4. [Table 3 (MATH GPT-4 row)] The relative-gain table contains an arithmetic error: GPT-4’s MATH accuracy is 68 versus a best-alternative 75, but the table lists a +3.03% gain rather than a -9.33% loss. Please correct the computation and re-verify all entries in Tables 3 and 4; this matters because the abstract’s “21% on MATH” is based on a relative gain from the same computation.
minor comments (6)
  1. [Section 1, contribution 2] The text reads “on the MATH dataset by XYZ”; this is an unresolved placeholder that must be filled with the actual reported value.
  2. [Section 8.3] The cross-reference “table (ref)” should be replaced with the specific table number (Table 6).
  3. [Limitations section] The sentences about ARC and ACRE contain an unrendered citation “( ??)”; please add the relevant references.
  4. [Abstract] The phrase “serve both as both students and teachers” contains a duplicated “both”; please revise.
  5. [Section 5.3] The evaluation section should state whether the 200-problem samples are a fixed random seed, how ties in human annotation were resolved, and whether inter-annotator agreement was measured; with exact-answer math problems this is a minor reporting gap, but it affects reproducibility.
  6. [Appendix tables] The repeated caption “Only scenarios with positive Z (gains) are reported” is inaccurate because negative z values are displayed as dashes in the same tables; rephrase to something like “Positive-z rows are indicated in bold.”

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical prompting comparison against external benchmarks, and its acknowledged solution-leakage limitation is a validity threat, not a circular derivation.

full rationale

This paper does not contain a formal derivation whose conclusions are equivalent to its assumptions by construction. Trace-of-Thought Prompting is an empirical prompting scheme: a teacher model generates decomposition steps and a student model solves the original question conditioned on those steps. There are no fitted parameters, no quantities defined in terms of the outcome being predicted, and no load-bearing self-citation chain. The reported accuracies are measured on external benchmarks (GSM8K and MATH) with human annotation against dataset labels, so the headline gains are not defined in terms of the teacher-generated steps themselves. The paper's own Limitations section flags a real concern: 'too strong of a prompt may encourage the teacher model to solve the problem in order to confirm its own reasoning, leading to inadvertent data contamination,' and Section 8.2 shows a delegation step containing the answer ('Acknowledge the function's constant value: **? = 3**'). However, this is a threat to the interpretation that the gains come from distilled reasoning rather than answer copying; it is not circularity in the derivation sense, because the measured accuracies would still be well-defined observations even if leakage occurred. No step in the paper reduces its claimed results to its inputs or to a self-citation, so the appropriate finding is no significant circularity.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The method rests on domain assumptions about decomposability and teacher fidelity rather than on fitted parameters. The main risk is that the teacher's steps may leak the answer, which the paper acknowledges.

assumptions (3)
  • domain assumption Problems such as GSM8K and MATH can be decomposed into a sequence of interpretable steps.
    Section 4.1 formalizes q -> {s1,...,sn}; the method presupposes that decomposition preserves enough information for the student to solve.
  • domain assumption The teacher model follows the instruction 'Do not solve the problem' and generates steps that do not leak the answer.
    The delegation prompt in Table 1 includes 'Do not solve the problem,' but Section 9 notes this may not hold, causing solution diffusion.
  • domain assumption The 200-sample subsets are representative of the full GSM8K and MATH benchmarks.
    Section 5.1 samples n=200 from each dataset without describing seed or stratification.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Trace-of-Thought Prompting: Investigating Prompt-Based Knowledge Distillation Through Question Decomposition." pith.science (2026). https://pith.science/paper/MZGYOO24

@misc{pith2026250420946,
  author       = {Pith},
  title        = {Pith review of: Trace-of-Thought Prompting: Investigating Prompt-Based Knowledge Distillation Through Question Decomposition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MZGYOO24}},
  note         = {Machine review of arXiv:2504.20946}
}
read the original abstract

Knowledge distillation allows smaller neural networks to emulate the performance of larger, teacher models with reduced computational demands. Traditional methods for Large Language Models (LLMs) often necessitate extensive fine-tuning, which limits their accessibility. To address this, we introduce Trace-of-Thought Prompting, a novel framework designed to distill critical reasoning capabilities from high-resource teacher models (over 8 billion parameters) to low-resource student models (up to 8 billion parameters). This approach leverages problem decomposition to enhance interpretability and facilitate human-in-the-loop interventions. Empirical evaluations on the GSM8K and MATH datasets show that student models achieve accuracy gains of up to 113% on GSM8K and 21% on MATH, with significant improvements particularly notable in smaller models like Llama 2 and Zephyr. Our results suggest a promising pathway for open-source, low-resource models to eventually serve both as both students and teachers, potentially reducing our reliance on high-resource, proprietary models.

Figures

Figures reproduced from arXiv: 2504.20946 by the authors.

Figure 1
Figure 1. A Visual Depiction of our Trace-of-Thought [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Visual depiction of the methods employed during experimentation. Trace-of-Thought provides a novel [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Relative accuracy changes with Trace-of [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 1 canonical work pages

  1. [1]

    Palaash Agrawal, Shavak Vasania, and Cheston Tan. 2024. http://arxiv.org/abs/2402.01805 Can llms perform structured graph reasoning?

    arXiv 2024

  2. [2]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litwin...

    arXiv 2020

  3. [3]

    Edward Y. Chang. 2024. http://arxiv.org/abs/2402.06634 Socrasynth: Multi-llm reasoning with conditional statistics

    arXiv 2024

  4. [4]

    Banghao Chen, Zhaofeng Zhang, Nicolas Langrené, and Shengxin Zhu. 2024 a . http://arxiv.org/abs/2310.14735 Unleashing the potential of prompt engineering: a comprehensive review

    arXiv 2024

  5. [5]

    Pei Chen, Boran Han, and Shuai Zhang. 2024 b . http://arxiv.org/abs/2404.17729 Comm: Collaborative multi-agent, multi-reasoning-path prompting for complex problem solving

    arXiv 2024

  6. [6]

    Zixiang Chen, Yihe Deng, Huizhuo Yuan, Kaixuan Ji, and Quanquan Gu. 2024 c . http://arxiv.org/abs/2401.01335 Self-play fine-tuning converts weak language models to strong language models

    arXiv 2024

  7. [7]

    Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. 2021. http://arxiv.org/abs/2110.14168 Training verifiers to solve math word problems

    arXiv 2021

  8. [8]

    Aniket Didolkar, Anirudh Goyal, Nan Rosemary Ke, Siyuan Guo, Michal Valko, Timothy Lillicrap, Danilo Rezende, Yoshua Bengio, Michael Mozer, and Sanjeev Arora. 2024. http://arxiv.org/abs/2405.12205 Metacognitive capabilities of llms: An exploration in mathematical problem solving

    arXiv 2024

Show all 45 references
  1. [9]

    Dheeru Dua, Shivanshu Gupta, Sameer Singh, and Matt Gardner. 2022. http://arxiv.org/abs/2212.04092 Successive prompting for decomposing complex questions

    2022 arXiv

  2. [10]

    Anmol Goel, Nico Daheim, and Iryna Gurevych. 2024. http://arxiv.org/abs/2403.03029 Socratic reasoning improves positive text rewriting

    2024 arXiv

  3. [11]

    Yuxian Gu, Li Dong, Furu Wei, and Minlie Huang. 2024. http://arxiv.org/abs/2306.08543 Minillm: Knowledge distillation of large language models

    2024 arXiv

  4. [12]

    Suriya Gunasekar, Yi Zhang, Jyoti Aneja, Caio César Teodoro Mendes, Allie Del Giorno, Sivakanth Gopi, Mojan Javaheripi, Piero Kauffmann, Gustavo de Rosa, Olli Saarikivi, Adil Salim, Shital Shah, Harkirat Singh Behl, Xin Wang, Sébastien Bubeck, Ronen Eldan, Adam Tauman Kalai, Y...

    2023 arXiv

  5. [13]

    Griffiths, and Mengdi Wang

    Xudong Guo, Kaixuan Huang, Jiale Liu, Wenhui Fan, Natalia Vélez, Qingyun Wu, Huazheng Wang, Thomas L. Griffiths, and Mengdi Wang. 2024. http://arxiv.org/abs/2403.12482 Embodied llm agents learn to cooperate in organized teams

    2024 arXiv

  6. [14]

    Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. 2015. http://arxiv.org/abs/1503.02531 Distilling the knowledge in a neural network

    2015 arXiv

  7. [15]

    Gurusha Juneja, Subhabrata Dutta, and Tanmoy Chakraborty. 2024. http://arxiv.org/abs/2404.02255 LM ^ 2 : A simple society of language models solves complex reasoning

    2024 arXiv

  8. [16]

    Tushar Khot, Harsh Trivedi, Matthew Finlayson, Yao Fu, Kyle Richardson, Peter Clark, and Ashish Sabharwal. 2023. http://arxiv.org/abs/2210.02406 Decomposed prompting: A modular approach for solving complex tasks

    2023 arXiv

  9. [17]

    Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. 2023. http://arxiv.org/abs/2205.11916 Large language models are zero-shot reasoners

    2023 arXiv

  10. [18]

    Yann LeCun, L \'e on Bottou, Yoshua Bengio, and Patrick Haffner. 1998. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11):2278--2324

    1998

  11. [19]

    Guohao Li, Hasan Abed Al Kader Hammoud, Hani Itani, Dmitrii Khizbullin, and Bernard Ghanem. 2023. http://arxiv.org/abs/2303.17760 Camel: Communicative agents for "mind" exploration of large language model society

    2023 arXiv

  12. [20]

    Tianle Li, Ge Zhang, Quy Duc Do, Xiang Yue, and Wenhu Chen. 2024. http://arxiv.org/abs/2404.02060 Long-context llms struggle with long in-context learning

    2024 arXiv

  13. [21]

    Chengyuan Liu, Yangyang Kang, Fubang Zhao, Kun Kuang, Zhuoren Jiang, Changlong Sun, and Fei Wu. 2024. http://arxiv.org/abs/2403.06414 Evolving knowledge distillation with large language models and active learning

    2024 arXiv

  14. [22]

    Amirkeivan Mohtashami, Florian Hartmann, Sian Gooding, Lukas Zilka, Matt Sharifi, and Blaise Aguera y Arcas. 2024. http://arxiv.org/abs/2312.11441 Social learning: Towards collaborative learning with large language models

    2024 arXiv

  15. [23]

    Smith, and Mike Lewis

    Ofir Press, Muru Zhang, Sewon Min, Ludwig Schmidt, Noah A. Smith, and Mike Lewis. 2023. http://arxiv.org/abs/2210.03350 Measuring and narrowing the compositionality gap in language models

    2023 arXiv

  16. [24]

    Jingyuan Qi, Zhiyang Xu, Ying Shen, Minqian Liu, Di Jin, Qifan Wang, and Lifu Huang. 2023. http://arxiv.org/abs/2305.14999 The art of socratic questioning: Recursive thinking with large language models

    2023 arXiv

  17. [25]

    Pranab Sahoo, Ayush Kumar Singh, Sriparna Saha, Vinija Jain, Samrat Mondal, and Aman Chadha. 2024. http://arxiv.org/abs/2402.07927 A systematic survey of prompt engineering in large language models: Techniques and applications

    2024 arXiv

  18. [26]

    Kumar Shridhar, Alessandro Stolfo, and Mrinmaya Sachan. 2023. http://arxiv.org/abs/2212.00193 Distilling reasoning capabilities into smaller language models

    2023 arXiv

  19. [27]

    Simeng Sun, Yang Liu, Shuohang Wang, Chenguang Zhu, and Mohit Iyyer. 2023. http://arxiv.org/abs/2305.14564 Pearl: Prompting large language models to plan and execute actions over long documents

    2023 arXiv

  20. [28]

    Gemma Team, Thomas Mesnard, Cassidy Hardin, Robert Dadashi, Surya Bhupatiraju, Shreya Pathak, Laurent Sifre, Morgane Rivière, Mihir Sanjay Kale, Juliette Love, Pouya Tafti, Léonard Hussenot, Pier Giuseppe Sessa, Aakanksha Chowdhery, Adam Roberts, Aditya Barua, Alex Botev, Alex...

    2024 arXiv

  21. [29]

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023. http://arxiv.org/abs/2302.13971 Llama: ...

    2023 arXiv

  22. [30]

    Rush, and Thomas Wolf

    Lewis Tunstall, Edward Beeching, Nathan Lambert, Nazneen Rajani, Kashif Rasul, Younes Belkada, Shengyi Huang, Leandro von Werra, Clémentine Fourrier, Nathan Habib, Nathan Sarrazin, Omar Sanseviero, Alexander M. Rush, and Thomas Wolf. 2023. http://arxiv.org/abs/2310.16944 Zephy...

    2023 arXiv

  23. [31]

    Cangqing Wang, Yutian Yang, Ruisi Li, Dan Sun, Ruicong Cai, Yuzhu Zhang, Chengqian Fu, and Lillian Floyd. 2024. http://arxiv.org/abs/2404.04997 Adapting llms for efficient context processing through soft prompt compression

    2024 arXiv

  24. [32]

    Lei Wang, Wanyu Xu, Yihuai Lan, Zhiqiang Hu, Yunshi Lan, Roy Ka-Wei Lee, and Ee-Peng Lim. 2023 a . http://arxiv.org/abs/2305.04091 Plan-and-solve prompting: Improving zero-shot chain-of-thought reasoning by large language models

    2023 arXiv

  25. [33]

    Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2023 b . http://arxiv.org/abs/2203.11171 Self-consistency improves chain of thought reasoning in language models

    2023 arXiv

  26. [34]

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. 2023. http://arxiv.org/abs/2201.11903 Chain-of-thought prompting elicits reasoning in large language models

    2023 arXiv

  27. [35]

    Yexin Wu, Zhuosheng Zhang, and Hai Zhao. 2024. http://arxiv.org/abs/2403.19167 Mitigating misleading chain-of-thought reasoning with selective filtering

    2024 arXiv

  28. [36]

    Can Xu, Qingfeng Sun, Kai Zheng, Xiubo Geng, Pu Zhao, Jiazhan Feng, Chongyang Tao, and Daxin Jiang. 2023. http://arxiv.org/abs/2304.12244 Wizardlm: Empowering large language models to follow complex instructions

    2023 arXiv

  29. [37]

    Xiaohan Xu, Ming Li, Chongyang Tao, Tao Shen, Reynold Cheng, Jinyang Li, Can Xu, Dacheng Tao, and Tianyi Zhou. 2024. http://arxiv.org/abs/2402.13116 A survey on knowledge distillation of large language models

    2024 arXiv

  30. [38]

    Griffiths, Yuan Cao, and Karthik Narasimhan

    Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, and Karthik Narasimhan. 2023. http://arxiv.org/abs/2305.10601 Tree of thoughts: Deliberate problem solving with large language models

    2023 arXiv

  31. [39]

    Andy Zeng, Maria Attarian, Brian Ichter, Krzysztof Choromanski, Adrian Wong, Stefan Welker, Federico Tombari, Aveek Purohit, Michael Ryoo, Vikas Sindhwani, Johnny Lee, Vincent Vanhoucke, and Pete Florence. 2022. http://arxiv.org/abs/2204.00598 Socratic models: Composing zero-s...

    2022 arXiv

  32. [40]

    Yunxiang Zhang, Muhammad Khalifa, Lajanugen Logeswaran, Jaekyeom Kim, Moontae Lee, Honglak Lee, and Lu Wang. 2024. http://arxiv.org/abs/2404.17140 Small language models need strong verifiers to self-correct reasoning

    2024 arXiv

  33. [41]

    Chuanyang Zheng, Zhengying Liu, Enze Xie, Zhenguo Li, and Yu Li. 2023. http://arxiv.org/abs/2304.09797 Progressive-hint prompting improves reasoning in large language models

    2023 arXiv

  34. [42]

    Qihuang Zhong, Liang Ding, Juhua Liu, Bo Du, and Dacheng Tao. 2024. http://arxiv.org/abs/2208.10160 Panda: Prompt transfer meets knowledge distillation for efficient model adaptation

    2024 arXiv

  35. [43]

    Denny Zhou, Nathanael Schärli, Le Hou, Jason Wei, Nathan Scales, Xuezhi Wang, Dale Schuurmans, Claire Cui, Olivier Bousquet, Quoc Le, and Ed Chi. 2023. http://arxiv.org/abs/2205.10625 Least-to-most prompting enables complex reasoning in large language models

    2023 arXiv

  36. [44]

    URL: " 'urlintro :=

    ENTRY address author booktitle chapter edition editor howpublished institution journal key month note number organization pages publisher school series title type volume year eprint doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before...

  37. [45]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.